Three years after launching its custom Maia chips, Microsoft is making the processors its primary inference system for its products, following in the footsteps of AWS and Google Cloud as custom chips become big business.
Microsoft said Monday that its Maia 200 chips were now in production and published the first details of its predecessor, Maia 300, alongside its plans to scale its Maia deployments to multi-gigawatt scale.
CEO Satya Nadella described the move as its “biggest bet yet on our own silicon” and said: “We are already deploying Maia 200 across the US, Europe, and Australia, delivering up to 1.7x more tokens per watt across experiences in Teams and Copilot.”
While Microsoft remains a large customer for Nvidia and its GPUs, its embrace of custom chips, manufactured by TSMC, follows similar moves by AWS and Google, which both run services on and sell their Trainium/Graviton and TPU chips.
